|
Rastor posted:Oh no doubt there are test machines in labs, I agree there. But stuff like being unstable under heavy compiling loads is gonna have to be ironed out before the production order is placed. Oh you sweet summer child
|
#
?
Jun 4, 2017 21:16
|
|
|
|
| # ? Apr 18, 2024 15:35 |
|
|
Oh dear:  Every new news item keeps making me flipflop between AMD and Intel. --edit: I love how that dumb spreadsheet linked in the Phoronix article asks for mainboard but not which CPU version. Might as well turn out to be an issue with binning, AMD being a little too optimistic about the higher end models or some poo poo like that. Combat Pretzel fucked around with this message at 21:24 on Jun 4, 2017 |
|
#
?
Jun 4, 2017 21:22
|
|
|
Combat Pretzel posted:Oh dear:  although if AMD keeps dropping prices I'll buy a cheap (if buggy) 1700/X370 system just to mess around with it until the HEDT picture clears up
|
|
#
?
Jun 4, 2017 21:25
|
|
|
Malcolm XML posted:Oh you sweet summer child Without violating NDAs, I know what I'm talking about
|
|
#
?
Jun 4, 2017 21:27
|
|
|
Combat Pretzel posted:Oh dear: What news article?
|
|
#
?
Jun 4, 2017 21:49
|
|
|
havenwaters posted:If you mean that thing where Rise of the Tomb Raider was running worse on amd cpus compared to intel cpus Crystal Dynamics just patched their game to fix that late last month.
|
|
#
?
Jun 4, 2017 21:51
|
|
|
ufarn posted:Ah, that's great to hear. Puts the ball in devs' court. That's a lovely place for a ball to be on this stuff.
|
|
#
?
Jun 5, 2017 00:48
|
|
|
Rastor posted:Without violating NDAs, I know what I'm talking about  : so do IIt is as if different providers and customers have different requirements even inside a cloud....
|
|
#
?
Jun 5, 2017 03:08
|
|
|
Everyone has different computers!
|
|
#
?
Jun 5, 2017 03:09
|
|
|
I don't know too much about virtualization but how transparently virtualizable are multi core multi GPU setups these days? Is there a big performance hit? Put another way, would it be possible to create a multi seat gaming setup with Threadripper and multiple Nividia cards?
|
|
#
?
Jun 5, 2017 08:02
|
|
|
shrike82 posted:I don't know too much about virtualization but how transparently virtualizable are multi core multi GPU setups these days? The performance hit with a hypervisor like KVM is negligible (1-5% in my experience). What you mentioned is possible, although there's currently a bug on the Ryzen platform that degrades performance.
|
|
#
?
Jun 5, 2017 08:25
|
|
|
eames, you spelunk the AMD forums, right? Have they finally acknowledged the problem with nested page tables?
|
|
#
?
Jun 5, 2017 09:45
|
|
|
Sadly no, they tout AGESA 1.0.0.6 as the fix for GPU passthrough but that only solves IOMMU groups. It might be a KVM bug (I found a bug report from 2008) but at the moment we have to assume like neither party is looking into it.  https://www.reddit.com/r/VFIO/ has good amount on info on virtualization and passthrough. e: this is the latest statement by the KVM devs https://lists.linuxfoundation.org/pipermail/iommu/2017-May/021690.html eames fucked around with this message at 10:11 on Jun 5, 2017 |
|
#
?
Jun 5, 2017 10:01
|
|
|
Well, crap. Let's hope that the only thing keeping them from addressing this particular problem is purely a case of EPYC/Threadripper crunch.
|
|
#
?
Jun 5, 2017 10:21
|
|
|
Strikes me as odd that they'd launch a brand new, heavily multithreaded server platform with such a virtualization bug in place. I guess PCIe passthrough is still pretty niche so they assume customers will leave NPT enabled. 
|
|
#
?
Jun 5, 2017 10:28
|
|
|
https://twitter.com/JokerReview/status/870764544532021248
|
|
#
?
Jun 6, 2017 09:58
|
|
|
Paul MaudDib posted:This is all about economies of scale. What's the savings from removing a few extra discretes and side-chips in mass-production quantities? Not much. How much does it cost to set up the pick-and-place machine for new boards, and track an entire other second lineup? At least some. How many are they going to sell? Well, either very few, or their entire stock - in which case you're now sold out for months until you can ramp production (by which time it may well be over). Necro quoting, with an interesting link that needs salt. https://videocardz.com/70162/amd-and-nvidia-preparing-graphics-cards-for-cryptocurrency-mining
|
|
#
?
Jun 6, 2017 13:45
|
|
|
I have literally owned an entire (terrible) laptop smaller than that.
|
|
#
?
Jun 6, 2017 17:26
|
|
|
It's also a photoshop.
|
|
#
?
Jun 6, 2017 17:40
|
|
|
PerrineClostermann posted:It's also a photoshop. Are you sure?
|
|
#
?
Jun 6, 2017 17:49
|
|
|
SinineSiil posted:Are you sure? The worst (best?) part is you can't be totally sure these days.
|
|
#
?
Jun 6, 2017 17:50
|
|
|
SinineSiil posted:Are you sure? What, you think it's actually possible that there's a picture of "Threadripper II" and it's the size of a motherboard? Also I can tell from the pixels and having seen quite a few shops in my time 
|
|
#
?
Jun 6, 2017 18:32
|
|
|
PerrineClostermann posted:What, you think it's actually possible that there's a picture of "Threadripper II" and it's the size of a motherboard? MoC = Motherboard on a Chip.  edit: and yes I know it's fake. 
|
|
#
?
Jun 6, 2017 18:35
|
|
|
PerrineClostermann posted:What, you think it's actually possible that there's a picture of "Threadripper II" and it's the size of a motherboard? spasticColon posted:edit: and yes I know it's fake. Sininu fucked around with this message at 18:54 on Jun 6, 2017 |
|
#
?
Jun 6, 2017 18:51
|
|
|
Does arctic silver come in pints?
|
|
#
?
Jun 6, 2017 19:33
|
|
|
GRINDCORE MEGGIDO posted:Necro quoting, with an interesting link that needs salt. edit: I'm gonna take this to the GPU thread
|
|
#
?
Jun 6, 2017 19:37
|
|
|
It *appears* that Samsung, working with GloFo and IBM, may beat Intel to market with EUV. IBM is touting 40% perf increase or 75% reduced power consumption from the new GAAFET transistor design vs commercial 10nm FinFET. https://www.theverge.com/circuitbreaker/2017/6/5/15740500/ibm-5nm-gaafet-transistor-faster-more-efficient https://arstechnica.com/gadgets/2017/06/ibm-5nm-chip/ http://wccftech.com/ibm-gf-5nm/ This is more bad news for Intel, I think. I can't remember if they had arrived at their next step after FinFET yet. Last I heard, they were going to switch to InGaAs at 7nm and ride FinFET all the way down to 5nm, or something like that. edit: Now the question remains: Can AMD GPU-side survive long enough to get whatever comes after Navi onto this new 5nm process? XD SwissArmyDruid fucked around with this message at 20:05 on Jun 6, 2017 |
|
#
?
Jun 6, 2017 19:54
|
|
|
Rumor claims entry-level 16core with SMT is $849: https://www.techpowerup.com/234114/amds-entry-level-16-core-32-thread-threadripper-to-reportedly-cost-usd-849
|
|
#
?
Jun 6, 2017 21:37
|
|
|
...how is that loving _entry level_? Isn't entry level supposed to be some 10-core part with SMT, clocked up to 3.4 with boost to 3.8?
|
|
#
?
Jun 6, 2017 22:50
|
|
|
SwissArmyDruid posted:...how is that loving _entry level_? Isn't entry level supposed to be some 10-core part with SMT, clocked up to 3.4 with boost to 3.8? I think it means the cheapest 16-core, not the cheapest out of any threadripper.
|
|
#
?
Jun 6, 2017 22:56
|
|
|
If there is a criticism that I have continually levelled at AMD, it is that they appear to be hell-bent on leaving as much money on the table as possible. That a 16-core Threadripper part of quite reasonable clock, might retail for _only_ $849? This implies that Threadripper, rather than commanding an HEDT price premium, as Intel has normalized over the past decade, will be a step up from R7 merely in the same way that one steps up from R5 -> R7. And I think that this is undercutting Intel WAY TOO HARD, and that AMD would benefit more in the short-term while not harming their long-term by pricing closer to Intel, but moving down whenever Intel moves. Either that, or AMD is burning cash to achieve lock-in on people who are increasingly turning against X299/Kaby Lake-X. Third Rule of Acquisition: Never spend more for an acquisition than you have to. SwissArmyDruid fucked around with this message at 10:00 on Jun 7, 2017 |
|
#
?
Jun 6, 2017 23:16
|
|
|
The other thing is that people infer quality from price. If something is 10% cheaper it's a good deal. If it's 30% cheaper it's a lower-tier product.
|
|
#
?
Jun 6, 2017 23:30
|
|
|
That's a pretty good price, especially if they're unlocked for overclocking
|
|
#
?
Jun 6, 2017 23:57
|
|
|
SwissArmyDruid posted:If there is a criticism that I have continually levelled at AMD, it is that they appear to be hell-bent on leaving as much money on the table as possible. AMD needs market share just as much as revenue.
|
|
#
?
Jun 7, 2017 00:42
|
|
|
Malcolm XML posted:AMD needs market share just as much as revenue. Yeah they really need to build market and mind share at this stage given their current HEDT market % is what.....0? also it could very well be true that making the ripper chips is really cheap and they are getting really high yields so even at half the price they are still turning a nice profit while also making people stop and go "gently caress 16 cores for that much I cant say no"
|
|
#
?
Jun 7, 2017 00:50
|
|
|
Scarecow posted:Yeah they really need to build market and mind share at this stage given their current HEDT market % is what.....0? also it could very well be true that making the ripper chips is really cheap and they are getting really high yields so even at half the price they are still turning a nice profit while also making people stop and go "gently caress 16 cores for that much I cant say no" Totally, 32 threads for $850 is enough for me to turn my 5820k into the worlds most overpowered NAS
|
|
#
?
Jun 7, 2017 00:51
|
|
|
I wonder what motherboards are going to look like to handle that many PCIe lanes out. More focus on EATX and EEB? Just all PCIe x16 at single-slot spacing to give good flexibility? This would actually be a really interesting design to use as a basis for GPU compute servers (although the last 4 lanes to get an even x16 per card would have been nice). Supermicro has some rack chassis with flexible risers in them I think. All you need then is a mezzanine card implementation of your GPU and you can engineer something that can compete with NVIDIA on rack density (like 1-2U). I think I remember NVIDIA having water-cooled Tesla servers now? That would be really effective for AMD too, you could easily make it a 1U with everything under water. Is Threadripper capable of multi-socket operation? Paul MaudDib fucked around with this message at 02:16 on Jun 7, 2017 |
|
#
?
Jun 7, 2017 01:06
|
|
|
Paul MaudDib posted:
I don't think so, IIRC the multi socket Epyc boards have 64 lanes dedicated to inter chip communication out of the 128 available.
|
|
#
?
Jun 7, 2017 01:31
|
|
|
the PCBs are going to be pretty dang thick for that many pins, if you can still find an image of LGA3647 laid horizontally you'd have an idea of how thicc of a pcb that sort of thing needs
|
|
#
?
Jun 7, 2017 02:15
|
|
|
|
| # ? Apr 18, 2024 15:35 |
|
|
.
sincx fucked around with this message at 05:50 on Mar 23, 2021 |
|
#
?
Jun 7, 2017 02:44
|
|
